Segmentation is a main problem in computer vision. It attempts to supply primitives for higher-level processes of interpretation. A multithreshold approach has already given acceptable results. This paper presents a PROLOG implementation of a multithreshold system. The system uses the technique of recursive partition of two-dimensional images in quadrants to form maximal blocks; the data structure generated by this partition algorithm is known as a quadtree. The algorithms implemented concern image representation, connected regions, identification, and contour detection. The improvement of algorithms in a logic language rather than in a procedural one appeared to be of particular interest. The main reasons for this are implementation facility and the natural use of logic programming in the field of knowledge-based systems. In fact, a knowledge-based approach seems a reasonable solution to problems of assisted image understanding processes.